通信学报 ›› 2022, Vol. 43 ›› Issue (6): 223-234.doi: 10.11959/j.issn.1000-436x.2022114

• 学术通信 • 上一篇    下一篇

车联网云边协同计算场景下的多目标优化卸载决策

朱思峰1, 蔡江昊1, 柴争义2, 孙恩林1   

  1. 1 天津城建大学计算机与信息工程学院，天津 300384
    2 天津工业大学计算机科学与技术学院，天津 300387
  • 修回日期:2022-05-09 出版日期:2022-06-01 发布日期:2022-06-01
  • 作者简介:朱思峰（1975- ），男，河南周口人，博士，天津城建大学教授，主要研究方向为边缘计算、人工智能算法及应用等
    蔡江昊（1998- ），男，湖北武汉人，天津城建大学硕士生，主要研究方向为边缘计算、人工智能算法等
    柴争义（1976- ），男，陕西渭南人，博士，天津工业大学教授，主要研究方向为智能物联网、边缘计算等
    孙恩林（1996- ），男，河北邢台人，天津城建大学硕士生，主要研究方向为边缘计算、人工智能算法等
  • 基金资助:
    国家自然科学基金资助项目(61972456);天津市自然科学基金资助项目(20JCYBJC00140);泛网无线通信教育部重点实验室（BUPT）开放课题基金资助项目(KFKT-2020101)

Multi-objective optimal offloading decision for cloud-edge collaborative computing scenario in Internet of vehicles

Sifeng ZHU1, Jianghao CAI1, Zhengyi CHAI2, Enlin SUN1   

  1. 1 School of Computer and Information Engineering, Tianjin Chengjian University, Tianjin 300384, China
    2 School of Computer Science ＆Technology, Tiangong University, Tianjin 300387, China
  • Revised:2022-05-09 Online:2022-06-01 Published:2022-06-01
  • Supported by:
    The National Natural Science Foundation of China(61972456);The Natural Science Foundation of Tianjin(20JCYBJC00140);The Open Project Fund of the Key Laboratory of Universal Wireless Communications (BUPT) of the Minis-try of Education(KFKT-2020101)

在线阅读

70

PDF下载

589

摘要：

目的：车联网场景下的计算任务对时延非常敏感，需要云边协同计算来满足这类需求。而车联网场景下车辆快速移动的特点使得常规的云边协同模型无法适用。本文结合车联网场景特有的车对车通信技术和边缘缓存技术，探索适用于车联网场景的云边协同计算卸载模型。

方法：针对车联网云边协同计算场景下如何高效地进行服务卸载并同时考虑服务的卸载决策以及边缘服务器和云服务器的协同资源分配问题，设计了基于云边协同的车辆计算网络架构，在该架构下，车载终端、云服务器和边缘服务器都可以提供计算服务；通过对缓存任务进行分类并将缓存策略引入车联网场景，依次设计了缓存模型、时延模型、能耗模型、服务质量模型以及多目标优化问题模型，将任务最大卸载时延引入服务质量模型；给出了一种基于改进的多目标优化免疫算法（MOIA）的卸载决策方案，该算法是一种多目标演化类算法，主要通过结合免疫思想和参考点策略实现对多目标问题的优化。

结果：最后，通过对比实验验证了所提卸载决策方案的有效性。实验结果表明，在满足最大卸载时延情况下本文提出的计算卸载模型能够应对不同需求的任务，具有较好的适应性。本文设计模型中卸载时延主要由七部分组成：任务下载服务应用所需的缓存时延、任务从车辆上传到边缘服务器的上传时延、任务从边缘服务器上传到云服务器的上传时延、任务所需的执行时延、任务在服务器端所需的排队时延、任务通过服务器进行跨区域传输所需的传输时延和任务通过基于车对车通信技术进行传输所需的传输时延。在对通信策略和缓存策略的实验中，可以看出本文中各部分时延有较为紧密的关联关系。对缓存策略效果的实验是通过取消一半可缓存的边缘缓存服务应用实现（MOIA-C），结果表明MOIA-C方案的卸载总时延和缓存时延较MOIA方案分别增加了35.88%和196.85%，这是由于可缓存服务应用数量的降低，卸载方案更倾向于将任务卸载到具有全部服务应用且性能更高的云服务器上，导致任务从边缘服务器上传到云服务器的上传时延和排队时延有所增长、执行时延有所降低，系统能量消耗下降，服务质量指标增加。通信策略采用基于服务器和基于车对车通信技术的混合传输方式，对通信策略的实验是通过取消基于车对车技术的通信方式实现（MOIA-S），结果表明MOIA-S方案的卸载总时延和通信总时延较MOIA方案分别增加了58.45%和433.33%，这是由于单独使用服务器进行任务传输会带来极大的带宽压力，为降低任务跨区域传输所带来的带宽压力，卸载方案更倾向于将任务卸载到云服务器执行，导致服务应用的缓存时延和任务的处理时延有所下降，而排队时延有所上升，系统能量消耗下降，服务质量指标增加。

结论：本文基于边缘缓存技术和车联网场景特有的车对车通信技术提出了一种自适应的服务缓存和任务卸载策略，在保证服务质量的基础上有效降低了车载任务的卸载总时延和车辆的能量消耗，可为车联网场景具有高时延敏感性任务提供了更为优质的服务。

关键词: 车联网, 云边协同, 卸载决策, 边缘缓存, 多目标优化免疫算法

Abstract:

Objectives:Computing tasks in Internet of vehicles are very sensitive to offloading delay, cloud-edge collaborative computing is required to meet such requirements. However,the characteristics of fast movement of vehicles in the Internet of vehicles make the conventional cloud-edge collaborative model not applicable. Combined with vehicle-to-vehicle communication technology and edge caching technology, this paper explores a cloud-edge collaborative computing offloading model suitable for The Internet of vehicles.

Methods:Aiming at the problem that in the cloud-edge collaborative computing scenario of the Internet of vehicles, it is a challenging problem how to efficiently offload services, and simultaneously consider the offloading decisions of services with the collaborative resource allocation of edge servers and cloud servers, a vehicle computing network architecture based on cloud-edge collaboration was designed. In this architecture, vehicle terminals, cloud servers and edge servers could provide computing services. The cache strategy was introduced into the scenario of Internet of vehicles by classifying cache tasks. The cache model, delay model, energy consumption model, quality of service model and multi-objective optimization model were designed successively,the maximum unload delay of tasks is introduced into the quality of service model. An improved multi-objective optimization immune algorithm（MOIA）was proposed for offloading decision making, the algorithm is a multi-objective evolutionary algorithm, mainly through the combination of immune thought and reference point strategy to achieve the optimization of multi-objective problems.

Results:Finally,the effectiveness of the proposed offloading decision scheme was verified by comparative experiments. Experimental results show that the computational offloading model proposed in this paper can cope with tasks with different requirements and has good adaptability under the condition of meeting the maximum offloading delay. Offloading delay in this design model is mainly composed of seven parts: The cache delay of service application required by task downloading from server, the uploading delay of task uploading from vehicle to edge server,the uploading delay of task uploading from edge server to cloud server, the execution delay required by task,the queuing delay required by task on server, the transmission delay required for tasks to be transmitted across regions through the server and the transmission delay required for tasks to be transmitted through vehicle-to-vehicle communication. In the experiment of communication strategy and cache strategy, it can be seen that each part of the delay in this paper has a relatively close relationship.The effect of cache strategy is tested by canceling half of cacheable edge cache service applications(MOIA-C).The results show that the total offload delay and cache delay of MOIA-C scheme increase 35.88% and 196.85% respectively compared with MOIA scheme, which is due to the decrease in the number of cacheable service applications. The scheme is more inclined to offload tasks to the cloud server that caches all service applications and has higher performance. As a result, the uploading delay of tasks from edge server to cloud server and the queuing delay of tasks on the server increase, the execution delay decreases,the system energy consumption decreases,and the service quality index increases.The communication strategy adopts the hybrid transmission mode based on server communication and vehicle-to-vehicle communication.The experiment of the communication strategy is realized by canceling the communication mode based on vehicle-to-vehicle technology (MOIA-S). The results show that the total offloading delay and communication delay of MOIA-S scheme increased by 58.45% and 433.33% respectively compared with MOIA scheme. This is due to the extreme bandwidth strain of using only the server to transport tasks. In order to reduce the bandwidth pressure caused by cross-region task transmission,the scheme tends to offload the task to the cloud server.Therefore,the cache delay of service application and the processing delay of the task decrease, the queuing delay increases, the system energy consumption decreases,and the quality of service index increases.

Conclusions:Based on vehicle-to-vehicle communication technology and edge caching technology, this paper proposes an adaptive service caching and task offloading strategy,which can effectively reduce the total delay of vehicles tasks and the energy consumption of vehicles while ensuring the quality of service, and provide better service for high-delay-sensitive tasks in Internet of vehicles scenarios.

Key words: Internet of vehicles, cloud-edge collaboration, offloading decision, edge cache, multi-objective optimization immune algorithm

中图分类号: 

No Suggested Reading articles found!